基于反射与透射图像结合的烟叶自动分级研究(英文)  

Grading Tobacco Leaves Based on a Combination of Reflectance and Transmittance Images

作　　者：刘华波[1] 贺立源[1] 马文杰[2] 

机构地区：[1]华中农业大学资源与环境学院,湖北武汉430070 [2]青岛农业大学艺术与传媒学院,山东青岛266109

出　　处：《应用基础与工程科学学报》2009年第3期343-350,共8页Journal of Basic Science and Engineering

基　　金：This work is partially supported by National key Technology R&D program(No.2006BAD10A1304).

摘　　要：人工进行烟叶外观分级需由目视和手感同时进行,而现有的利用计算机视觉技术对烟叶进行自动识别时,通常只利用了烟叶的反射图像,对烟叶的内部品质特征难以进行有效的模拟.本研究在自行设计制作的灯箱内拍摄同一烟叶的反射和透射二幅图像,通过数字图像处理技术自动提取烟叶的颜色和几何特征.运用人工神经网络建模,解决分级过程中的复杂非线性关系.实验结果证明,透射图像特征能够反映烟叶的某些内部分级因素,结合二类图像特征进行的识别效率明显高于使用某一类图像单独进行识别.同时,建模时所需特征参数大大减少,提高了运行效率和适应性.Inspectors Currently, automated assess the quality of tobacco leaves by sight and touch. grading of tobacco leaves uses only reflected light in a vision machine. Such methods make it difficult to identify the internal attributes of leaves. Human inspectors use touch to discern these internal qualities. In this study, we designed a novel image acquisition device which could capture two kinds of images from both reflected and transmitted light for each tobacco leaf. Color and geometrical features extracted from both types of images were input into artificial neural networks and used to grade flue-cured tobacco leaves. The results showed that the combination of both types of images led to a better evaluation than either single image type.

关 键 词：烟叶 计算机视觉 透射图像 质量分级 

分 类 号：S572[农业科学—烟草工业] TP274[农业科学—作物学]